10 #include "nodestack.h"
13 #include "clockvector.h"
14 #include "cyclegraph.h"
17 #include "threads-model.h"
18 #include "bugmessage.h"
22 #define INITIAL_THREAD_ID 0
25 * Structure for holding small ModelChecker members that should be snapshotted
27 struct model_snapshot_members {
28 model_snapshot_members() :
29 /* First thread created will have id INITIAL_THREAD_ID */
30 next_thread_id(INITIAL_THREAD_ID),
31 used_sequence_numbers(0),
34 failed_promise(false),
35 hard_failed_promise(false),
36 too_many_reads(false),
37 no_valid_reads(false),
38 bad_synchronization(false),
43 ~model_snapshot_members() {
44 for (unsigned int i = 0; i < bugs.size(); i++)
49 unsigned int next_thread_id;
50 modelclock_t used_sequence_numbers;
51 ModelAction *next_backtrack;
52 SnapVector<bug_message *> bugs;
54 bool hard_failed_promise;
57 /** @brief Incorrectly-ordered synchronization was made */
58 bool bad_synchronization;
65 /** @brief Constructor */
66 ModelExecution::ModelExecution(ModelChecker *m,
67 const struct model_params *params,
69 NodeStack *node_stack) :
74 thread_map(2), /* We'll always need at least 2 threads */
76 condvar_waiters_map(),
82 thrd_last_fence_release(),
83 node_stack(node_stack),
84 priv(new struct model_snapshot_members()),
85 mo_graph(new CycleGraph())
87 /* Initialize a model-checker thread, for special ModelActions */
88 model_thread = new Thread(get_next_id()); // L: Create model thread
89 add_thread(model_thread); // L: Add model thread to scheduler
90 scheduler->register_engine(this);
91 node_stack->register_engine(this);
94 /** @brief Destructor */
95 ModelExecution::~ModelExecution()
97 for (unsigned int i = 0; i < get_num_threads(); i++)
98 delete get_thread(int_to_id(i));
100 for (unsigned int i = 0; i < promises.size(); i++)
107 int ModelExecution::get_execution_number() const
109 return model->get_execution_number();
112 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
114 action_list_t *tmp = hash->get(ptr);
116 tmp = new action_list_t();
122 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
124 SnapVector<action_list_t> *tmp = hash->get(ptr);
126 tmp = new SnapVector<action_list_t>();
132 action_list_t * ModelExecution::get_actions_on_obj(void * obj, thread_id_t tid) const
134 SnapVector<action_list_t> *wrv = obj_thrd_map.get(obj);
137 unsigned int thread=id_to_int(tid);
138 if (thread < wrv->size())
139 return &(*wrv)[thread];
144 /** @return a thread ID for a new Thread */
145 thread_id_t ModelExecution::get_next_id()
147 return priv->next_thread_id++;
150 /** @return the number of user threads created during this execution */
151 unsigned int ModelExecution::get_num_threads() const
153 return priv->next_thread_id;
156 /** @return a sequence number for a new ModelAction */
157 modelclock_t ModelExecution::get_next_seq_num()
159 return ++priv->used_sequence_numbers;
163 * @brief Should the current action wake up a given thread?
165 * @param curr The current action
166 * @param thread The thread that we might wake up
167 * @return True, if we should wake up the sleeping thread; false otherwise
169 bool ModelExecution::should_wake_up(const ModelAction *curr, const Thread *thread) const
171 const ModelAction *asleep = thread->get_pending();
172 /* Don't allow partial RMW to wake anyone up */
175 /* Synchronizing actions may have been backtracked */
176 if (asleep->could_synchronize_with(curr))
178 /* All acquire/release fences and fence-acquire/store-release */
179 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
181 /* Fence-release + store can awake load-acquire on the same location */
182 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
183 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
184 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
190 void ModelExecution::wake_up_sleeping_actions(ModelAction *curr)
192 for (unsigned int i = 0; i < get_num_threads(); i++) {
193 Thread *thr = get_thread(int_to_id(i));
194 if (scheduler->is_sleep_set(thr)) {
195 if (should_wake_up(curr, thr))
196 /* Remove this thread from sleep set */
197 scheduler->remove_sleep(thr);
202 /** @brief Alert the model-checker that an incorrectly-ordered
203 * synchronization was made */
204 void ModelExecution::set_bad_synchronization()
206 priv->bad_synchronization = true;
209 /** @brief Alert the model-checker that an incorrectly-ordered
210 * synchronization was made */
211 void ModelExecution::set_bad_sc_read()
213 priv->bad_sc_read = true;
216 bool ModelExecution::assert_bug(const char *msg)
218 priv->bugs.push_back(new bug_message(msg));
220 if (isfeasibleprefix()) {
227 /** @return True, if any bugs have been reported for this execution */
228 bool ModelExecution::have_bug_reports() const
230 return priv->bugs.size() != 0;
233 SnapVector<bug_message *> * ModelExecution::get_bugs() const
239 * Check whether the current trace has triggered an assertion which should halt
242 * @return True, if the execution should be aborted; false otherwise
244 bool ModelExecution::has_asserted() const
246 return priv->asserted;
250 * Trigger a trace assertion which should cause this execution to be halted.
251 * This can be due to a detected bug or due to an infeasibility that should
254 void ModelExecution::set_assert()
256 priv->asserted = true;
260 * Check if we are in a deadlock. Should only be called at the end of an
261 * execution, although it should not give false positives in the middle of an
262 * execution (there should be some ENABLED thread).
264 * @return True if program is in a deadlock; false otherwise
266 bool ModelExecution::is_deadlocked() const
268 bool blocking_threads = false;
269 for (unsigned int i = 0; i < get_num_threads(); i++) {
270 thread_id_t tid = int_to_id(i);
273 Thread *t = get_thread(tid);
274 if (!t->is_model_thread() && t->get_pending())
275 blocking_threads = true;
277 return blocking_threads;
281 * @brief Check if we are yield-blocked
283 * A program can be "yield-blocked" if all threads are ready to execute a
286 * @return True if the program is yield-blocked; false otherwise
288 bool ModelExecution::is_yieldblocked() const
290 if (!params->yieldblock)
293 for (unsigned int i = 0; i < get_num_threads(); i++) {
294 thread_id_t tid = int_to_id(i);
295 Thread *t = get_thread(tid);
296 if (t->get_pending() && t->get_pending()->is_yield())
303 * Check if this is a complete execution. That is, have all thread completed
304 * execution (rather than exiting because sleep sets have forced a redundant
307 * @return True if the execution is complete.
309 bool ModelExecution::is_complete_execution() const
311 if (is_yieldblocked())
313 for (unsigned int i = 0; i < get_num_threads(); i++)
314 if (is_enabled(int_to_id(i)))
320 * @brief Find the last fence-related backtracking conflict for a ModelAction
322 * This function performs the search for the most recent conflicting action
323 * against which we should perform backtracking, as affected by fence
324 * operations. This includes pairs of potentially-synchronizing actions which
325 * occur due to fence-acquire or fence-release, and hence should be explored in
326 * the opposite execution order.
328 * @param act The current action
329 * @return The most recent action which conflicts with act due to fences
331 ModelAction * ModelExecution::get_last_fence_conflict(ModelAction *act) const
333 /* Only perform release/acquire fence backtracking for stores */
334 if (!act->is_write())
337 /* Find a fence-release (or, act is a release) */
338 ModelAction *last_release;
339 if (act->is_release())
342 last_release = get_last_fence_release(act->get_tid());
346 /* Skip past the release */
347 const action_list_t *list = &action_trace;
348 action_list_t::const_reverse_iterator rit;
349 for (rit = list->rbegin(); rit != list->rend(); rit++)
350 if (*rit == last_release)
352 ASSERT(rit != list->rend());
357 * load --sb-> fence-acquire */
358 ModelVector<ModelAction *> acquire_fences(get_num_threads(), NULL);
359 ModelVector<ModelAction *> prior_loads(get_num_threads(), NULL);
360 bool found_acquire_fences = false;
361 for ( ; rit != list->rend(); rit++) {
362 ModelAction *prev = *rit;
363 if (act->same_thread(prev))
366 int tid = id_to_int(prev->get_tid());
368 if (prev->is_read() && act->same_var(prev)) {
369 if (prev->is_acquire()) {
370 /* Found most recent load-acquire, don't need
371 * to search for more fences */
372 if (!found_acquire_fences)
375 prior_loads[tid] = prev;
378 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
379 found_acquire_fences = true;
380 acquire_fences[tid] = prev;
384 ModelAction *latest_backtrack = NULL;
385 for (unsigned int i = 0; i < acquire_fences.size(); i++)
386 if (acquire_fences[i] && prior_loads[i])
387 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
388 latest_backtrack = acquire_fences[i];
389 return latest_backtrack;
393 * @brief Find the last backtracking conflict for a ModelAction
395 * This function performs the search for the most recent conflicting action
396 * against which we should perform backtracking. This primary includes pairs of
397 * synchronizing actions which should be explored in the opposite execution
400 * @param act The current action
401 * @return The most recent action which conflicts with act
403 ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
405 switch (act->get_type()) {
407 /* Only seq-cst fences can (directly) cause backtracking */
408 if (!act->is_seqcst())
413 ModelAction *ret = NULL;
415 /* linear search: from most recent to oldest */
416 action_list_t *list = obj_map.get(act->get_location());
417 action_list_t::reverse_iterator rit;
418 for (rit = list->rbegin(); rit != list->rend(); rit++) {
419 ModelAction *prev = *rit;
422 if (prev->could_synchronize_with(act)) {
428 ModelAction *ret2 = get_last_fence_conflict(act);
438 case ATOMIC_TRYLOCK: {
439 /* linear search: from most recent to oldest */
440 action_list_t *list = obj_map.get(act->get_location());
441 action_list_t::reverse_iterator rit;
442 for (rit = list->rbegin(); rit != list->rend(); rit++) {
443 ModelAction *prev = *rit;
444 if (act->is_conflicting_lock(prev))
449 case ATOMIC_UNLOCK: {
450 /* linear search: from most recent to oldest */
451 action_list_t *list = obj_map.get(act->get_location());
452 action_list_t::reverse_iterator rit;
453 for (rit = list->rbegin(); rit != list->rend(); rit++) {
454 ModelAction *prev = *rit;
455 if (!act->same_thread(prev) && prev->is_failed_trylock())
461 /* linear search: from most recent to oldest */
462 action_list_t *list = obj_map.get(act->get_location());
463 action_list_t::reverse_iterator rit;
464 for (rit = list->rbegin(); rit != list->rend(); rit++) {
465 ModelAction *prev = *rit;
466 if (!act->same_thread(prev) && prev->is_failed_trylock())
468 if (!act->same_thread(prev) && prev->is_notify())
474 case ATOMIC_NOTIFY_ALL:
475 case ATOMIC_NOTIFY_ONE: {
476 /* linear search: from most recent to oldest */
477 action_list_t *list = obj_map.get(act->get_location());
478 action_list_t::reverse_iterator rit;
479 for (rit = list->rbegin(); rit != list->rend(); rit++) {
480 ModelAction *prev = *rit;
481 if (!act->same_thread(prev) && prev->is_wait())
492 /** This method finds backtracking points where we should try to
493 * reorder the parameter ModelAction against.
495 * @param the ModelAction to find backtracking points for.
497 void ModelExecution::set_backtracking(ModelAction *act)
499 Thread *t = get_thread(act);
500 ModelAction *prev = get_last_conflict(act);
504 Node *node = prev->get_node()->get_parent();
506 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
507 int low_tid, high_tid;
508 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
509 low_tid = id_to_int(act->get_tid());
510 high_tid = low_tid + 1;
513 high_tid = get_num_threads();
516 for (int i = low_tid; i < high_tid; i++) {
517 thread_id_t tid = int_to_id(i);
519 /* Make sure this thread can be enabled here. */
520 if (i >= node->get_num_threads())
523 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
524 /* Don't backtrack into a point where the thread is disabled or sleeping. */
525 if (node->enabled_status(tid) != THREAD_ENABLED)
528 /* Check if this has been explored already */
529 if (node->has_been_explored(tid))
532 /* See if fairness allows */
533 if (params->fairwindow != 0 && !node->has_priority(tid)) {
535 for (int t = 0; t < node->get_num_threads(); t++) {
536 thread_id_t tother = int_to_id(t);
537 if (node->is_enabled(tother) && node->has_priority(tother)) {
546 /* See if CHESS-like yield fairness allows */
547 if (params->yieldon) {
549 for (int t = 0; t < node->get_num_threads(); t++) {
550 thread_id_t tother = int_to_id(t);
551 if (node->is_enabled(tother) && node->has_priority_over(tid, tother)) {
560 /* Cache the latest backtracking point */
561 set_latest_backtrack(prev);
563 /* If this is a new backtracking point, mark the tree */
564 if (!node->set_backtrack(tid))
566 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
567 id_to_int(prev->get_tid()),
568 id_to_int(t->get_id()));
577 * @brief Cache the a backtracking point as the "most recent", if eligible
579 * Note that this does not prepare the NodeStack for this backtracking
580 * operation, it only caches the action on a per-execution basis
582 * @param act The operation at which we should explore a different next action
583 * (i.e., backtracking point)
584 * @return True, if this action is now the most recent backtracking point;
587 bool ModelExecution::set_latest_backtrack(ModelAction *act)
589 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
590 priv->next_backtrack = act;
597 * Returns last backtracking point. The model checker will explore a different
598 * path for this point in the next execution.
599 * @return The ModelAction at which the next execution should diverge.
601 ModelAction * ModelExecution::get_next_backtrack()
603 ModelAction *next = priv->next_backtrack;
604 priv->next_backtrack = NULL;
609 * Processes a read model action.
610 * @param curr is the read model action to process.
611 * @return True if processing this read updates the mo_graph.
613 bool ModelExecution::process_read(ModelAction *curr)
615 Node *node = curr->get_node();
617 bool updated = false;
618 switch (node->get_read_from_status()) {
619 case READ_FROM_PAST: {
620 const ModelAction *rf = node->get_read_from_past();
623 mo_graph->startChanges();
625 ASSERT(!is_infeasible());
626 if (!check_recency(curr, rf)) {
627 if (node->increment_read_from()) {
628 mo_graph->rollbackChanges();
631 priv->too_many_reads = true;
635 updated = r_modification_order(curr, rf);
637 mo_graph->commitChanges();
638 mo_check_promises(curr, true);
641 case READ_FROM_PROMISE: {
642 Promise *promise = curr->get_node()->get_read_from_promise();
643 if (promise->add_reader(curr))
644 priv->failed_promise = true;
645 curr->set_read_from_promise(promise);
646 mo_graph->startChanges();
647 if (!check_recency(curr, promise))
648 priv->too_many_reads = true;
649 updated = r_modification_order(curr, promise);
650 mo_graph->commitChanges();
653 case READ_FROM_FUTURE: {
654 /* Read from future value */
655 struct future_value fv = node->get_future_value();
656 Promise *promise = new Promise(this, curr, fv);
657 curr->set_read_from_promise(promise);
658 promises.push_back(promise);
659 mo_graph->startChanges();
660 updated = r_modification_order(curr, promise);
661 mo_graph->commitChanges();
667 get_thread(curr)->set_return_value(curr->get_return_value());
673 * Processes a lock, trylock, or unlock model action. @param curr is
674 * the read model action to process.
676 * The try lock operation checks whether the lock is taken. If not,
677 * it falls to the normal lock operation case. If so, it returns
680 * The lock operation has already been checked that it is enabled, so
681 * it just grabs the lock and synchronizes with the previous unlock.
683 * The unlock operation has to re-enable all of the threads that are
684 * waiting on the lock.
686 * @return True if synchronization was updated; false otherwise
688 bool ModelExecution::process_mutex(ModelAction *curr)
690 std::mutex *mutex = curr->get_mutex();
691 struct std::mutex_state *state = NULL;
694 state = mutex->get_state();
696 switch (curr->get_type()) {
697 case ATOMIC_TRYLOCK: {
698 bool success = !state->locked;
699 curr->set_try_lock(success);
701 get_thread(curr)->set_return_value(0);
704 get_thread(curr)->set_return_value(1);
706 //otherwise fall into the lock case
708 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
709 assert_bug("Lock access before initialization");
710 state->locked = get_thread(curr);
711 ModelAction *unlock = get_last_unlock(curr);
712 //synchronize with the previous unlock statement
713 if (unlock != NULL) {
714 synchronize(unlock, curr);
720 case ATOMIC_UNLOCK: {
721 /* wake up the other threads */
722 for (unsigned int i = 0; i < get_num_threads(); i++) {
723 Thread *t = get_thread(int_to_id(i));
724 Thread *curr_thrd = get_thread(curr);
725 if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
729 /* unlock the lock - after checking who was waiting on it */
730 state->locked = NULL;
732 if (!curr->is_wait())
733 break; /* The rest is only for ATOMIC_WAIT */
735 /* Should we go to sleep? (simulate spurious failures) */
736 if (curr->get_node()->get_misc() == 0) {
737 get_safe_ptr_action(&condvar_waiters_map, curr->get_location())->push_back(curr);
739 scheduler->sleep(get_thread(curr));
743 case ATOMIC_NOTIFY_ALL: {
744 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
745 //activate all the waiting threads
746 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
747 scheduler->wake(get_thread(*rit));
752 case ATOMIC_NOTIFY_ONE: {
753 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
754 int wakeupthread = curr->get_node()->get_misc();
755 action_list_t::iterator it = waiters->begin();
756 advance(it, wakeupthread);
757 scheduler->wake(get_thread(*it));
769 * @brief Check if the current pending promises allow a future value to be sent
771 * It is unsafe to pass a future value back if there exists a pending promise Pr
774 * reader --exec-> Pr --exec-> writer
776 * If such Pr exists, we must save the pending future value until Pr is
779 * @param writer The operation which sends the future value. Must be a write.
780 * @param reader The operation which will observe the value. Must be a read.
781 * @return True if the future value can be sent now; false if it must wait.
783 bool ModelExecution::promises_may_allow(const ModelAction *writer,
784 const ModelAction *reader) const
786 for (int i = promises.size() - 1; i >= 0; i--) {
787 ModelAction *pr = promises[i]->get_reader(0);
788 //reader is after promise...doesn't cross any promise
791 //writer is after promise, reader before...bad...
799 * @brief Add a future value to a reader
801 * This function performs a few additional checks to ensure that the future
802 * value can be feasibly observed by the reader
804 * @param writer The operation whose value is sent. Must be a write.
805 * @param reader The read operation which may read the future value. Must be a read.
807 void ModelExecution::add_future_value(const ModelAction *writer, ModelAction *reader)
809 /* Do more ambitious checks now that mo is more complete */
810 if (!mo_may_allow(writer, reader))
813 Node *node = reader->get_node();
815 /* Find an ancestor thread which exists at the time of the reader */
816 Thread *write_thread = get_thread(writer);
817 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
818 write_thread = write_thread->get_parent();
820 struct future_value fv = {
821 writer->get_write_value(),
822 writer->get_seq_number() + params->maxfuturedelay,
823 write_thread->get_id(),
825 if (node->add_future_value(fv))
826 set_latest_backtrack(reader);
830 * Process a write ModelAction
831 * @param curr The ModelAction to process
832 * @param work The work queue, for adding fixup work
833 * @return True if the mo_graph was updated or promises were resolved
835 bool ModelExecution::process_write(ModelAction *curr, work_queue_t *work)
837 /* Readers to which we may send our future value */
838 ModelVector<ModelAction *> send_fv;
840 const ModelAction *earliest_promise_reader;
841 bool updated_promises = false;
843 bool updated_mod_order = w_modification_order(curr, &send_fv);
844 Promise *promise = pop_promise_to_resolve(curr);
847 earliest_promise_reader = promise->get_reader(0);
848 updated_promises = resolve_promise(curr, promise, work);
850 earliest_promise_reader = NULL;
852 for (unsigned int i = 0; i < send_fv.size(); i++) {
853 ModelAction *read = send_fv[i];
855 /* Don't send future values to reads after the Promise we resolve */
856 if (!earliest_promise_reader || *read < *earliest_promise_reader) {
857 /* Check if future value can be sent immediately */
858 if (promises_may_allow(curr, read)) {
859 add_future_value(curr, read);
861 futurevalues.push_back(PendingFutureValue(curr, read));
866 /* Check the pending future values */
867 for (int i = (int)futurevalues.size() - 1; i >= 0; i--) {
868 struct PendingFutureValue pfv = futurevalues[i];
869 if (promises_may_allow(pfv.writer, pfv.reader)) {
870 add_future_value(pfv.writer, pfv.reader);
871 futurevalues.erase(futurevalues.begin() + i);
875 mo_graph->commitChanges();
876 mo_check_promises(curr, false);
878 get_thread(curr)->set_return_value(VALUE_NONE);
879 return updated_mod_order || updated_promises;
883 * Process a fence ModelAction
884 * @param curr The ModelAction to process
885 * @return True if synchronization was updated
887 bool ModelExecution::process_fence(ModelAction *curr)
890 * fence-relaxed: no-op
891 * fence-release: only log the occurence (not in this function), for
892 * use in later synchronization
893 * fence-acquire (this function): search for hypothetical release
895 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
897 bool updated = false;
898 if (curr->is_acquire()) {
899 action_list_t *list = &action_trace;
900 action_list_t::reverse_iterator rit;
901 /* Find X : is_read(X) && X --sb-> curr */
902 for (rit = list->rbegin(); rit != list->rend(); rit++) {
903 ModelAction *act = *rit;
906 if (act->get_tid() != curr->get_tid())
908 /* Stop at the beginning of the thread */
909 if (act->is_thread_start())
911 /* Stop once we reach a prior fence-acquire */
912 if (act->is_fence() && act->is_acquire())
916 /* read-acquire will find its own release sequences */
917 if (act->is_acquire())
920 /* Establish hypothetical release sequences */
921 rel_heads_list_t release_heads;
922 get_release_seq_heads(curr, act, &release_heads);
923 for (unsigned int i = 0; i < release_heads.size(); i++)
924 synchronize(release_heads[i], curr);
925 if (release_heads.size() != 0)
933 * @brief Process the current action for thread-related activity
935 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
936 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
937 * synchronization, etc. This function is a no-op for non-THREAD actions
938 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
940 * @param curr The current action
941 * @return True if synchronization was updated or a thread completed
943 bool ModelExecution::process_thread_action(ModelAction *curr)
945 bool updated = false;
947 switch (curr->get_type()) {
948 case THREAD_CREATE: {
949 thrd_t *thrd = (thrd_t *)curr->get_location();
950 struct thread_params *params = (struct thread_params *)curr->get_value();
951 Thread *th = new Thread(get_next_id(), thrd, params->func, params->arg, get_thread(curr));
953 th->set_creation(curr);
954 /* Promises can be satisfied by children */
955 for (unsigned int i = 0; i < promises.size(); i++) {
956 Promise *promise = promises[i];
957 if (promise->thread_is_available(curr->get_tid()))
958 promise->add_thread(th->get_id());
962 case PTHREAD_CREATE: {
963 thrd_t *thrd = (thrd_t *)curr->get_location();
964 struct pthread_params *params = (struct pthread_params *)curr->get_value();
965 Thread *th = new Thread(get_next_id(), thrd, params->func, params->arg, get_thread(curr));
967 th->set_creation(curr);
968 /* Promises can be satisfied by children */
969 for (unsigned int i = 0; i < promises.size(); i++) {
970 Promise *promise = promises[i];
971 if (promise->thread_is_available(curr->get_tid()))
972 promise->add_thread(th->get_id());
977 Thread *blocking = curr->get_thread_operand();
978 ModelAction *act = get_last_action(blocking->get_id());
979 synchronize(act, curr);
980 updated = true; /* trigger rel-seq checks */
984 break; // WL: to be add (modified)
985 Thread *blocking = curr->get_thread_operand();
986 ModelAction *act = get_last_action(blocking->get_id());
987 synchronize(act, curr);
988 updated = true; /* trigger rel-seq checks */
991 case THREAD_FINISH: {
992 Thread *th = get_thread(curr);
993 /* Wake up any joining threads */
994 for (unsigned int i = 0; i < get_num_threads(); i++) {
995 Thread *waiting = get_thread(int_to_id(i));
996 if (waiting->waiting_on() == th &&
997 waiting->get_pending()->is_thread_join())
998 scheduler->wake(waiting);
1001 /* Completed thread can't satisfy promises */
1002 for (unsigned int i = 0; i < promises.size(); i++) {
1003 Promise *promise = promises[i];
1004 if (promise->thread_is_available(th->get_id()))
1005 if (promise->eliminate_thread(th->get_id()))
1006 priv->failed_promise = true;
1008 updated = true; /* trigger rel-seq checks */
1011 case THREAD_START: {
1012 check_promises(curr->get_tid(), NULL, curr->get_cv());
1023 * @brief Process the current action for release sequence fixup activity
1025 * Performs model-checker release sequence fixups for the current action,
1026 * forcing a single pending release sequence to break (with a given, potential
1027 * "loose" write) or to complete (i.e., synchronize). If a pending release
1028 * sequence forms a complete release sequence, then we must perform the fixup
1029 * synchronization, mo_graph additions, etc.
1031 * @param curr The current action; must be a release sequence fixup action
1032 * @param work_queue The work queue to which to add work items as they are
1035 void ModelExecution::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1037 const ModelAction *write = curr->get_node()->get_relseq_break();
1038 struct release_seq *sequence = pending_rel_seqs.back();
1039 pending_rel_seqs.pop_back();
1041 ModelAction *acquire = sequence->acquire;
1042 const ModelAction *rf = sequence->rf;
1043 const ModelAction *release = sequence->release;
1047 ASSERT(release->same_thread(rf));
1049 if (write == NULL) {
1051 * @todo Forcing a synchronization requires that we set
1052 * modification order constraints. For instance, we can't allow
1053 * a fixup sequence in which two separate read-acquire
1054 * operations read from the same sequence, where the first one
1055 * synchronizes and the other doesn't. Essentially, we can't
1056 * allow any writes to insert themselves between 'release' and
1060 /* Must synchronize */
1061 if (!synchronize(release, acquire))
1064 /* Propagate the changed clock vector */
1065 propagate_clockvector(acquire, work_queue);
1067 /* Break release sequence with new edges:
1068 * release --mo--> write --mo--> rf */
1069 mo_graph->addEdge(release, write);
1070 mo_graph->addEdge(write, rf);
1073 /* See if we have realized a data race */
1078 * Initialize the current action by performing one or more of the following
1079 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1080 * in the NodeStack, manipulating backtracking sets, allocating and
1081 * initializing clock vectors, and computing the promises to fulfill.
1083 * @param curr The current action, as passed from the user context; may be
1084 * freed/invalidated after the execution of this function, with a different
1085 * action "returned" its place (pass-by-reference)
1086 * @return True if curr is a newly-explored action; false otherwise
1088 bool ModelExecution::initialize_curr_action(ModelAction **curr)
1090 ModelAction *newcurr;
1092 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1093 newcurr = process_rmw(*curr);
1096 if (newcurr->is_rmw())
1097 compute_promises(newcurr);
1103 (*curr)->set_seq_number(get_next_seq_num());
1105 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1107 /* First restore type and order in case of RMW operation */
1108 if ((*curr)->is_rmwr())
1109 newcurr->copy_typeandorder(*curr);
1111 ASSERT((*curr)->get_location() == newcurr->get_location());
1112 newcurr->copy_from_new(*curr);
1114 /* Discard duplicate ModelAction; use action from NodeStack */
1117 /* Always compute new clock vector */
1118 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1121 return false; /* Action was explored previously */
1125 /* Always compute new clock vector */
1126 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1128 /* Assign most recent release fence */
1129 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1132 * Perform one-time actions when pushing new ModelAction onto
1135 if (newcurr->is_write())
1136 compute_promises(newcurr);
1137 else if (newcurr->is_relseq_fixup())
1138 compute_relseq_breakwrites(newcurr);
1139 else if (newcurr->is_wait())
1140 newcurr->get_node()->set_misc_max(2);
1141 else if (newcurr->is_notify_one()) {
1142 newcurr->get_node()->set_misc_max(get_safe_ptr_action(&condvar_waiters_map, newcurr->get_location())->size());
1144 return true; /* This was a new ModelAction */
1149 * @brief Establish reads-from relation between two actions
1151 * Perform basic operations involved with establishing a concrete rf relation,
1152 * including setting the ModelAction data and checking for release sequences.
1154 * @param act The action that is reading (must be a read)
1155 * @param rf The action from which we are reading (must be a write)
1157 * @return True if this read established synchronization
1160 bool ModelExecution::read_from(ModelAction *act, const ModelAction *rf)
1163 ASSERT(rf->is_write());
1165 act->set_read_from(rf);
1166 if (act->is_acquire()) {
1167 rel_heads_list_t release_heads;
1168 get_release_seq_heads(act, act, &release_heads);
1169 int num_heads = release_heads.size();
1170 for (unsigned int i = 0; i < release_heads.size(); i++)
1171 if (!synchronize(release_heads[i], act))
1173 return num_heads > 0;
1179 * @brief Synchronizes two actions
1181 * When A synchronizes with B (or A --sw-> B), B inherits A's clock vector.
1182 * This function performs the synchronization as well as providing other hooks
1183 * for other checks along with synchronization.
1185 * @param first The left-hand side of the synchronizes-with relation
1186 * @param second The right-hand side of the synchronizes-with relation
1187 * @return True if the synchronization was successful (i.e., was consistent
1188 * with the execution order); false otherwise
1190 bool ModelExecution::synchronize(const ModelAction *first, ModelAction *second)
1192 if (*second < *first) {
1193 set_bad_synchronization();
1196 check_promises(first->get_tid(), second->get_cv(), first->get_cv());
1197 return second->synchronize_with(first);
1201 * Check promises and eliminate potentially-satisfying threads when a thread is
1202 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1203 * no longer satisfy a promise generated from that thread.
1205 * @param blocker The thread on which a thread is waiting
1206 * @param waiting The waiting thread
1208 void ModelExecution::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1210 for (unsigned int i = 0; i < promises.size(); i++) {
1211 Promise *promise = promises[i];
1212 if (!promise->thread_is_available(waiting->get_id()))
1214 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1215 ModelAction *reader = promise->get_reader(j);
1216 if (reader->get_tid() != blocker->get_id())
1218 if (promise->eliminate_thread(waiting->get_id())) {
1219 /* Promise has failed */
1220 priv->failed_promise = true;
1222 /* Only eliminate the 'waiting' thread once */
1230 * @brief Check whether a model action is enabled.
1232 * Checks whether an operation would be successful (i.e., is a lock already
1233 * locked, or is the joined thread already complete).
1235 * For yield-blocking, yields are never enabled.
1237 * @param curr is the ModelAction to check whether it is enabled.
1238 * @return a bool that indicates whether the action is enabled.
1240 bool ModelExecution::check_action_enabled(ModelAction *curr) {
1241 if (curr->is_lock()) {
1242 std::mutex *lock = curr->get_mutex();
1243 struct std::mutex_state *state = lock->get_state();
1246 } else if (curr->is_thread_join()) {
1247 Thread *blocking = curr->get_thread_operand();
1248 if (!blocking->is_complete()) {
1249 thread_blocking_check_promises(blocking, get_thread(curr));
1252 } else if (params->yieldblock && curr->is_yield()) {
1260 * This is the heart of the model checker routine. It performs model-checking
1261 * actions corresponding to a given "current action." Among other processes, it
1262 * calculates reads-from relationships, updates synchronization clock vectors,
1263 * forms a memory_order constraints graph, and handles replay/backtrack
1264 * execution when running permutations of previously-observed executions.
1266 * @param curr The current action to process
1267 * @return The ModelAction that is actually executed; may be different than
1270 ModelAction * ModelExecution::check_current_action(ModelAction *curr)
1273 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1274 bool newly_explored = initialize_curr_action(&curr);
1278 wake_up_sleeping_actions(curr);
1280 /* Compute fairness information for CHESS yield algorithm */
1281 if (params->yieldon) {
1282 curr->get_node()->update_yield(scheduler);
1285 /* Add the action to lists before any other model-checking tasks */
1286 if (!second_part_of_rmw)
1287 add_action_to_lists(curr);
1289 /* Build may_read_from set for newly-created actions */
1290 if (newly_explored && curr->is_read())
1291 build_may_read_from(curr);
1293 /* Initialize work_queue with the "current action" work */
1294 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1295 while (!work_queue.empty() && !has_asserted()) {
1296 WorkQueueEntry work = work_queue.front();
1297 work_queue.pop_front();
1299 switch (work.type) {
1300 case WORK_CHECK_CURR_ACTION: {
1301 ModelAction *act = work.action;
1302 bool update = false; /* update this location's release seq's */
1303 bool update_all = false; /* update all release seq's */
1305 if (process_thread_action(curr))
1308 if (act->is_read() && !second_part_of_rmw && process_read(act))
1311 if (act->is_write() && process_write(act, &work_queue))
1314 if (act->is_fence() && process_fence(act))
1317 if (act->is_mutex_op() && process_mutex(act))
1320 if (act->is_relseq_fixup())
1321 process_relseq_fixup(curr, &work_queue);
1324 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1326 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1329 case WORK_CHECK_RELEASE_SEQ:
1330 resolve_release_sequences(work.location, &work_queue);
1332 case WORK_CHECK_MO_EDGES: {
1333 /** @todo Complete verification of work_queue */
1334 ModelAction *act = work.action;
1335 bool updated = false;
1337 if (act->is_read()) {
1338 const ModelAction *rf = act->get_reads_from();
1339 const Promise *promise = act->get_reads_from_promise();
1341 if (r_modification_order(act, rf))
1343 if (act->is_seqcst()) {
1344 ModelAction *last_sc_write = get_last_seq_cst_write(act);
1345 if (last_sc_write != NULL && rf->happens_before(last_sc_write)) {
1349 } else if (promise) {
1350 if (r_modification_order(act, promise))
1354 if (act->is_write()) {
1355 if (w_modification_order(act, NULL))
1358 mo_graph->commitChanges();
1361 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1370 check_curr_backtracking(curr);
1371 set_backtracking(curr);
1375 void ModelExecution::check_curr_backtracking(ModelAction *curr)
1377 Node *currnode = curr->get_node();
1378 Node *parnode = currnode->get_parent();
1380 if ((parnode && !parnode->backtrack_empty()) ||
1381 !currnode->misc_empty() ||
1382 !currnode->read_from_empty() ||
1383 !currnode->promise_empty() ||
1384 !currnode->relseq_break_empty()) {
1385 set_latest_backtrack(curr);
1389 bool ModelExecution::promises_expired() const
1391 for (unsigned int i = 0; i < promises.size(); i++) {
1392 Promise *promise = promises[i];
1393 if (promise->get_expiration() < priv->used_sequence_numbers)
1400 * This is the strongest feasibility check available.
1401 * @return whether the current trace (partial or complete) must be a prefix of
1404 bool ModelExecution::isfeasibleprefix() const
1406 return pending_rel_seqs.size() == 0 && is_feasible_prefix_ignore_relseq();
1410 * Print disagnostic information about an infeasible execution
1411 * @param prefix A string to prefix the output with; if NULL, then a default
1412 * message prefix will be provided
1414 void ModelExecution::print_infeasibility(const char *prefix) const
1418 if (mo_graph->checkForCycles())
1419 ptr += sprintf(ptr, "[mo cycle]");
1420 if (priv->failed_promise || priv->hard_failed_promise)
1421 ptr += sprintf(ptr, "[failed promise]");
1422 if (priv->too_many_reads)
1423 ptr += sprintf(ptr, "[too many reads]");
1424 if (priv->no_valid_reads)
1425 ptr += sprintf(ptr, "[no valid reads-from]");
1426 if (priv->bad_synchronization)
1427 ptr += sprintf(ptr, "[bad sw ordering]");
1428 if (priv->bad_sc_read)
1429 ptr += sprintf(ptr, "[bad sc read]");
1430 if (promises_expired())
1431 ptr += sprintf(ptr, "[promise expired]");
1432 if (promises.size() != 0)
1433 ptr += sprintf(ptr, "[unresolved promise]");
1435 model_print("%s: %s", prefix ? prefix : "Infeasible", buf);
1439 * Returns whether the current completed trace is feasible, except for pending
1440 * release sequences.
1442 bool ModelExecution::is_feasible_prefix_ignore_relseq() const
1444 return !is_infeasible() && promises.size() == 0 && ! priv->failed_promise;
1449 * Check if the current partial trace is infeasible. Does not check any
1450 * end-of-execution flags, which might rule out the execution. Thus, this is
1451 * useful only for ruling an execution as infeasible.
1452 * @return whether the current partial trace is infeasible.
1454 bool ModelExecution::is_infeasible() const
1456 return mo_graph->checkForCycles() ||
1457 priv->no_valid_reads ||
1458 priv->too_many_reads ||
1459 priv->bad_synchronization ||
1460 priv->bad_sc_read ||
1461 priv->hard_failed_promise ||
1465 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1466 ModelAction * ModelExecution::process_rmw(ModelAction *act) {
1467 ModelAction *lastread = get_last_action(act->get_tid());
1468 lastread->process_rmw(act);
1469 if (act->is_rmw()) {
1470 if (lastread->get_reads_from())
1471 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1473 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1474 mo_graph->commitChanges();
1480 * A helper function for ModelExecution::check_recency, to check if the current
1481 * thread is able to read from a different write/promise for 'params.maxreads'
1482 * number of steps and if that write/promise should become visible (i.e., is
1483 * ordered later in the modification order). This helps model memory liveness.
1485 * @param curr The current action. Must be a read.
1486 * @param rf The write/promise from which we plan to read
1487 * @param other_rf The write/promise from which we may read
1488 * @return True if we were able to read from other_rf for params.maxreads steps
1490 template <typename T, typename U>
1491 bool ModelExecution::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1493 /* Need a different write/promise */
1494 if (other_rf->equals(rf))
1497 /* Only look for "newer" writes/promises */
1498 if (!mo_graph->checkReachable(rf, other_rf))
1501 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1502 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1503 action_list_t::reverse_iterator rit = list->rbegin();
1504 ASSERT((*rit) == curr);
1505 /* Skip past curr */
1508 /* Does this write/promise work for everyone? */
1509 for (int i = 0; i < params->maxreads; i++, rit++) {
1510 ModelAction *act = *rit;
1511 if (!act->may_read_from(other_rf))
1518 * Checks whether a thread has read from the same write or Promise for too many
1519 * times without seeing the effects of a later write/Promise.
1522 * 1) there must a different write/promise that we could read from,
1523 * 2) we must have read from the same write/promise in excess of maxreads times,
1524 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1525 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1527 * If so, we decide that the execution is no longer feasible.
1529 * @param curr The current action. Must be a read.
1530 * @param rf The ModelAction/Promise from which we might read.
1531 * @return True if the read should succeed; false otherwise
1533 template <typename T>
1534 bool ModelExecution::check_recency(ModelAction *curr, const T *rf) const
1536 if (!params->maxreads)
1539 //NOTE: Next check is just optimization, not really necessary....
1540 if (curr->get_node()->get_read_from_past_size() +
1541 curr->get_node()->get_read_from_promise_size() <= 1)
1544 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1545 int tid = id_to_int(curr->get_tid());
1546 ASSERT(tid < (int)thrd_lists->size());
1547 action_list_t *list = &(*thrd_lists)[tid];
1548 action_list_t::reverse_iterator rit = list->rbegin();
1549 ASSERT((*rit) == curr);
1550 /* Skip past curr */
1553 action_list_t::reverse_iterator ritcopy = rit;
1554 /* See if we have enough reads from the same value */
1555 for (int count = 0; count < params->maxreads; ritcopy++, count++) {
1556 if (ritcopy == list->rend())
1558 ModelAction *act = *ritcopy;
1559 if (!act->is_read())
1561 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1563 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1565 if (act->get_node()->get_read_from_past_size() +
1566 act->get_node()->get_read_from_promise_size() <= 1)
1569 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1570 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1571 if (should_read_instead(curr, rf, write))
1572 return false; /* liveness failure */
1574 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1575 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1576 if (should_read_instead(curr, rf, promise))
1577 return false; /* liveness failure */
1583 * @brief Updates the mo_graph with the constraints imposed from the current
1586 * Basic idea is the following: Go through each other thread and find
1587 * the last action that happened before our read. Two cases:
1589 * -# The action is a write: that write must either occur before
1590 * the write we read from or be the write we read from.
1591 * -# The action is a read: the write that that action read from
1592 * must occur before the write we read from or be the same write.
1594 * @param curr The current action. Must be a read.
1595 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1596 * @return True if modification order edges were added; false otherwise
1598 template <typename rf_type>
1599 bool ModelExecution::r_modification_order(ModelAction *curr, const rf_type *rf)
1601 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1604 ASSERT(curr->is_read());
1606 /* Last SC fence in the current thread */
1607 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1608 ModelAction *last_sc_write = NULL;
1609 if (curr->is_seqcst())
1610 last_sc_write = get_last_seq_cst_write(curr);
1612 /* Iterate over all threads */
1613 for (i = 0; i < thrd_lists->size(); i++) {
1614 /* Last SC fence in thread i */
1615 ModelAction *last_sc_fence_thread_local = NULL;
1616 if (int_to_id((int)i) != curr->get_tid())
1617 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1619 /* Last SC fence in thread i, before last SC fence in current thread */
1620 ModelAction *last_sc_fence_thread_before = NULL;
1621 if (last_sc_fence_local)
1622 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1624 /* Iterate over actions in thread, starting from most recent */
1625 action_list_t *list = &(*thrd_lists)[i];
1626 action_list_t::reverse_iterator rit;
1627 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1628 ModelAction *act = *rit;
1633 /* Don't want to add reflexive edges on 'rf' */
1634 if (act->equals(rf)) {
1635 if (act->happens_before(curr))
1641 if (act->is_write()) {
1642 /* C++, Section 29.3 statement 5 */
1643 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1644 *act < *last_sc_fence_thread_local) {
1645 added = mo_graph->addEdge(act, rf) || added;
1648 /* C++, Section 29.3 statement 4 */
1649 else if (act->is_seqcst() && last_sc_fence_local &&
1650 *act < *last_sc_fence_local) {
1651 added = mo_graph->addEdge(act, rf) || added;
1654 /* C++, Section 29.3 statement 6 */
1655 else if (last_sc_fence_thread_before &&
1656 *act < *last_sc_fence_thread_before) {
1657 added = mo_graph->addEdge(act, rf) || added;
1663 * Include at most one act per-thread that "happens
1666 if (act->happens_before(curr)) {
1667 if (act->is_write()) {
1668 added = mo_graph->addEdge(act, rf) || added;
1670 const ModelAction *prevrf = act->get_reads_from();
1671 const Promise *prevrf_promise = act->get_reads_from_promise();
1673 if (!prevrf->equals(rf))
1674 added = mo_graph->addEdge(prevrf, rf) || added;
1675 } else if (!prevrf_promise->equals(rf)) {
1676 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1685 * All compatible, thread-exclusive promises must be ordered after any
1686 * concrete loads from the same thread
1688 for (unsigned int i = 0; i < promises.size(); i++)
1689 if (promises[i]->is_compatible_exclusive(curr))
1690 added = mo_graph->addEdge(rf, promises[i]) || added;
1696 * Updates the mo_graph with the constraints imposed from the current write.
1698 * Basic idea is the following: Go through each other thread and find
1699 * the lastest action that happened before our write. Two cases:
1701 * (1) The action is a write => that write must occur before
1704 * (2) The action is a read => the write that that action read from
1705 * must occur before the current write.
1707 * This method also handles two other issues:
1709 * (I) Sequential Consistency: Making sure that if the current write is
1710 * seq_cst, that it occurs after the previous seq_cst write.
1712 * (II) Sending the write back to non-synchronizing reads.
1714 * @param curr The current action. Must be a write.
1715 * @param send_fv A vector for stashing reads to which we may pass our future
1716 * value. If NULL, then don't record any future values.
1717 * @return True if modification order edges were added; false otherwise
1719 bool ModelExecution::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
1721 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1724 ASSERT(curr->is_write());
1726 if (curr->is_seqcst()) {
1727 /* We have to at least see the last sequentially consistent write,
1728 so we are initialized. */
1729 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1730 if (last_seq_cst != NULL) {
1731 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1735 /* Last SC fence in the current thread */
1736 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1738 /* Iterate over all threads */
1739 for (i = 0; i < thrd_lists->size(); i++) {
1740 /* Last SC fence in thread i, before last SC fence in current thread */
1741 ModelAction *last_sc_fence_thread_before = NULL;
1742 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1743 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1745 /* Iterate over actions in thread, starting from most recent */
1746 action_list_t *list = &(*thrd_lists)[i];
1747 action_list_t::reverse_iterator rit;
1748 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1749 ModelAction *act = *rit;
1752 * 1) If RMW and it actually read from something, then we
1753 * already have all relevant edges, so just skip to next
1756 * 2) If RMW and it didn't read from anything, we should
1757 * whatever edge we can get to speed up convergence.
1759 * 3) If normal write, we need to look at earlier actions, so
1760 * continue processing list.
1762 if (curr->is_rmw()) {
1763 if (curr->get_reads_from() != NULL)
1771 /* C++, Section 29.3 statement 7 */
1772 if (last_sc_fence_thread_before && act->is_write() &&
1773 *act < *last_sc_fence_thread_before) {
1774 added = mo_graph->addEdge(act, curr) || added;
1779 * Include at most one act per-thread that "happens
1782 if (act->happens_before(curr)) {
1784 * Note: if act is RMW, just add edge:
1786 * The following edge should be handled elsewhere:
1787 * readfrom(act) --mo--> act
1789 if (act->is_write())
1790 added = mo_graph->addEdge(act, curr) || added;
1791 else if (act->is_read()) {
1792 //if previous read accessed a null, just keep going
1793 if (act->get_reads_from() == NULL) {
1794 added = mo_graph->addEdge(act->get_reads_from_promise(), curr) || added;
1796 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
1799 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1800 !act->same_thread(curr)) {
1801 /* We have an action that:
1802 (1) did not happen before us
1803 (2) is a read and we are a write
1804 (3) cannot synchronize with us
1805 (4) is in a different thread
1807 that read could potentially read from our write. Note that
1808 these checks are overly conservative at this point, we'll
1809 do more checks before actually removing the
1814 if (send_fv && thin_air_constraint_may_allow(curr, act) && check_coherence_promise(curr, act)) {
1815 if (!is_infeasible())
1816 send_fv->push_back(act);
1817 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
1818 add_future_value(curr, act);
1825 * All compatible, thread-exclusive promises must be ordered after any
1826 * concrete stores to the same thread, or else they can be merged with
1829 for (unsigned int i = 0; i < promises.size(); i++)
1830 if (promises[i]->is_compatible_exclusive(curr))
1831 added = mo_graph->addEdge(curr, promises[i]) || added;
1836 //This procedure uses cohere to prune future values that are
1837 //guaranteed to generate a coherence violation.
1839 //need to see if there is (1) a promise for thread write, (2)
1840 //the promise is sb before write, (3) the promise can only be
1841 //resolved by the thread read, and (4) the promise has same
1842 //location as read/write
1844 bool ModelExecution::check_coherence_promise(const ModelAction * write, const ModelAction *read) {
1845 thread_id_t write_tid=write->get_tid();
1846 for(unsigned int i = promises.size(); i>0; i--) {
1847 Promise *pr=promises[i-1];
1848 if (!pr->same_location(write))
1850 //the reading thread is the only thread that can resolve the promise
1851 if (pr->get_num_was_available_threads()==1 && pr->thread_was_available(read->get_tid())) {
1852 for(unsigned int j=0;j<pr->get_num_readers();j++) {
1853 ModelAction *prreader=pr->get_reader(j);
1854 //the writing thread reads from the promise before the write
1855 if (prreader->get_tid()==write_tid &&
1856 (*prreader)<(*write)) {
1857 if ((*read)>(*prreader)) {
1858 //check that we don't have a read between the read and promise
1859 //from the same thread as read
1861 for(const ModelAction *tmp=read;tmp!=prreader;) {
1862 tmp=tmp->get_node()->get_parent()->get_action();
1863 if (tmp->is_read() && tmp->same_thread(read)) {
1880 /** Arbitrary reads from the future are not allowed. Section 29.3
1881 * part 9 places some constraints. This method checks one result of constraint
1882 * constraint. Others require compiler support. */
1883 bool ModelExecution::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader) const
1885 if (!writer->is_rmw())
1888 if (!reader->is_rmw())
1891 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1892 if (search == reader)
1894 if (search->get_tid() == reader->get_tid() &&
1895 search->happens_before(reader))
1903 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1904 * some constraints. This method checks one the following constraint (others
1905 * require compiler support):
1907 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1908 * If X --hb-> Y, A --rf-> Y, and A --mo-> Z, then X should not read from Z.
1910 bool ModelExecution::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1912 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(reader->get_location());
1914 /* Iterate over all threads */
1915 for (i = 0; i < thrd_lists->size(); i++) {
1916 const ModelAction *write_after_read = NULL;
1918 /* Iterate over actions in thread, starting from most recent */
1919 action_list_t *list = &(*thrd_lists)[i];
1920 action_list_t::reverse_iterator rit;
1921 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1922 ModelAction *act = *rit;
1924 /* Don't disallow due to act == reader */
1925 if (!reader->happens_before(act) || reader == act)
1927 else if (act->is_write())
1928 write_after_read = act;
1929 else if (act->is_read() && act->get_reads_from() != NULL)
1930 write_after_read = act->get_reads_from();
1933 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
1940 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1941 * The ModelAction under consideration is expected to be taking part in
1942 * release/acquire synchronization as an object of the "reads from" relation.
1943 * Note that this can only provide release sequence support for RMW chains
1944 * which do not read from the future, as those actions cannot be traced until
1945 * their "promise" is fulfilled. Similarly, we may not even establish the
1946 * presence of a release sequence with certainty, as some modification order
1947 * constraints may be decided further in the future. Thus, this function
1948 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1949 * and a boolean representing certainty.
1951 * @param rf The action that might be part of a release sequence. Must be a
1953 * @param release_heads A pass-by-reference style return parameter. After
1954 * execution of this function, release_heads will contain the heads of all the
1955 * relevant release sequences, if any exists with certainty
1956 * @param pending A pass-by-reference style return parameter which is only used
1957 * when returning false (i.e., uncertain). Returns most information regarding
1958 * an uncertain release sequence, including any write operations that might
1959 * break the sequence.
1960 * @return true, if the ModelExecution is certain that release_heads is complete;
1963 bool ModelExecution::release_seq_heads(const ModelAction *rf,
1964 rel_heads_list_t *release_heads,
1965 struct release_seq *pending) const
1967 /* Only check for release sequences if there are no cycles */
1968 if (mo_graph->checkForCycles())
1971 for ( ; rf != NULL; rf = rf->get_reads_from()) {
1972 ASSERT(rf->is_write());
1974 if (rf->is_release())
1975 release_heads->push_back(rf);
1976 else if (rf->get_last_fence_release())
1977 release_heads->push_back(rf->get_last_fence_release());
1979 break; /* End of RMW chain */
1981 /** @todo Need to be smarter here... In the linux lock
1982 * example, this will run to the beginning of the program for
1984 /** @todo The way to be smarter here is to keep going until 1
1985 * thread has a release preceded by an acquire and you've seen
1988 /* acq_rel RMW is a sufficient stopping condition */
1989 if (rf->is_acquire() && rf->is_release())
1990 return true; /* complete */
1993 /* read from future: need to settle this later */
1995 return false; /* incomplete */
1998 if (rf->is_release())
1999 return true; /* complete */
2001 /* else relaxed write
2002 * - check for fence-release in the same thread (29.8, stmt. 3)
2003 * - check modification order for contiguous subsequence
2004 * -> rf must be same thread as release */
2006 const ModelAction *fence_release = rf->get_last_fence_release();
2007 /* Synchronize with a fence-release unconditionally; we don't need to
2008 * find any more "contiguous subsequence..." for it */
2010 release_heads->push_back(fence_release);
2012 int tid = id_to_int(rf->get_tid());
2013 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(rf->get_location());
2014 action_list_t *list = &(*thrd_lists)[tid];
2015 action_list_t::const_reverse_iterator rit;
2017 /* Find rf in the thread list */
2018 rit = std::find(list->rbegin(), list->rend(), rf);
2019 ASSERT(rit != list->rend());
2021 /* Find the last {write,fence}-release */
2022 for (; rit != list->rend(); rit++) {
2023 if (fence_release && *(*rit) < *fence_release)
2025 if ((*rit)->is_release())
2028 if (rit == list->rend()) {
2029 /* No write-release in this thread */
2030 return true; /* complete */
2031 } else if (fence_release && *(*rit) < *fence_release) {
2032 /* The fence-release is more recent (and so, "stronger") than
2033 * the most recent write-release */
2034 return true; /* complete */
2035 } /* else, need to establish contiguous release sequence */
2036 ModelAction *release = *rit;
2038 ASSERT(rf->same_thread(release));
2040 pending->writes.clear();
2042 bool certain = true;
2043 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
2044 if (id_to_int(rf->get_tid()) == (int)i)
2046 list = &(*thrd_lists)[i];
2048 /* Can we ensure no future writes from this thread may break
2049 * the release seq? */
2050 bool future_ordered = false;
2052 ModelAction *last = get_last_action(int_to_id(i));
2053 Thread *th = get_thread(int_to_id(i));
2054 if ((last && rf->happens_before(last)) ||
2057 future_ordered = true;
2059 ASSERT(!th->is_model_thread() || future_ordered);
2061 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2062 const ModelAction *act = *rit;
2063 /* Reach synchronization -> this thread is complete */
2064 if (act->happens_before(release))
2066 if (rf->happens_before(act)) {
2067 future_ordered = true;
2071 /* Only non-RMW writes can break release sequences */
2072 if (!act->is_write() || act->is_rmw())
2075 /* Check modification order */
2076 if (mo_graph->checkReachable(rf, act)) {
2077 /* rf --mo--> act */
2078 future_ordered = true;
2081 if (mo_graph->checkReachable(act, release))
2082 /* act --mo--> release */
2084 if (mo_graph->checkReachable(release, act) &&
2085 mo_graph->checkReachable(act, rf)) {
2086 /* release --mo-> act --mo--> rf */
2087 return true; /* complete */
2089 /* act may break release sequence */
2090 pending->writes.push_back(act);
2093 if (!future_ordered)
2094 certain = false; /* This thread is uncertain */
2098 release_heads->push_back(release);
2099 pending->writes.clear();
2101 pending->release = release;
2108 * An interface for getting the release sequence head(s) with which a
2109 * given ModelAction must synchronize. This function only returns a non-empty
2110 * result when it can locate a release sequence head with certainty. Otherwise,
2111 * it may mark the internal state of the ModelExecution so that it will handle
2112 * the release sequence at a later time, causing @a acquire to update its
2113 * synchronization at some later point in execution.
2115 * @param acquire The 'acquire' action that may synchronize with a release
2117 * @param read The read action that may read from a release sequence; this may
2118 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2119 * when 'acquire' is a fence-acquire)
2120 * @param release_heads A pass-by-reference return parameter. Will be filled
2121 * with the head(s) of the release sequence(s), if they exists with certainty.
2122 * @see ModelExecution::release_seq_heads
2124 void ModelExecution::get_release_seq_heads(ModelAction *acquire,
2125 ModelAction *read, rel_heads_list_t *release_heads)
2127 const ModelAction *rf = read->get_reads_from();
2128 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2129 sequence->acquire = acquire;
2130 sequence->read = read;
2132 if (!release_seq_heads(rf, release_heads, sequence)) {
2133 /* add act to 'lazy checking' list */
2134 pending_rel_seqs.push_back(sequence);
2136 snapshot_free(sequence);
2141 * @brief Propagate a modified clock vector to actions later in the execution
2144 * After an acquire operation lazily completes a release-sequence
2145 * synchronization, we must update all clock vectors for operations later than
2146 * the acquire in the execution order.
2148 * @param acquire The ModelAction whose clock vector must be propagated
2149 * @param work The work queue to which we can add work items, if this
2150 * propagation triggers more updates (e.g., to the modification order)
2152 void ModelExecution::propagate_clockvector(ModelAction *acquire, work_queue_t *work)
2154 /* Re-check all pending release sequences */
2155 work->push_back(CheckRelSeqWorkEntry(NULL));
2156 /* Re-check read-acquire for mo_graph edges */
2157 work->push_back(MOEdgeWorkEntry(acquire));
2159 /* propagate synchronization to later actions */
2160 action_list_t::reverse_iterator rit = action_trace.rbegin();
2161 for (; (*rit) != acquire; rit++) {
2162 ModelAction *propagate = *rit;
2163 if (acquire->happens_before(propagate)) {
2164 synchronize(acquire, propagate);
2165 /* Re-check 'propagate' for mo_graph edges */
2166 work->push_back(MOEdgeWorkEntry(propagate));
2172 * Attempt to resolve all stashed operations that might synchronize with a
2173 * release sequence for a given location. This implements the "lazy" portion of
2174 * determining whether or not a release sequence was contiguous, since not all
2175 * modification order information is present at the time an action occurs.
2177 * @param location The location/object that should be checked for release
2178 * sequence resolutions. A NULL value means to check all locations.
2179 * @param work_queue The work queue to which to add work items as they are
2181 * @return True if any updates occurred (new synchronization, new mo_graph
2184 bool ModelExecution::resolve_release_sequences(void *location, work_queue_t *work_queue)
2186 bool updated = false;
2187 SnapVector<struct release_seq *>::iterator it = pending_rel_seqs.begin();
2188 while (it != pending_rel_seqs.end()) {
2189 struct release_seq *pending = *it;
2190 ModelAction *acquire = pending->acquire;
2191 const ModelAction *read = pending->read;
2193 /* Only resolve sequences on the given location, if provided */
2194 if (location && read->get_location() != location) {
2199 const ModelAction *rf = read->get_reads_from();
2200 rel_heads_list_t release_heads;
2202 complete = release_seq_heads(rf, &release_heads, pending);
2203 for (unsigned int i = 0; i < release_heads.size(); i++)
2204 if (!acquire->has_synchronized_with(release_heads[i]))
2205 if (synchronize(release_heads[i], acquire))
2209 /* Propagate the changed clock vector */
2210 propagate_clockvector(acquire, work_queue);
2213 it = pending_rel_seqs.erase(it);
2214 snapshot_free(pending);
2220 // If we resolved promises or data races, see if we have realized a data race.
2227 * Performs various bookkeeping operations for the current ModelAction. For
2228 * instance, adds action to the per-object, per-thread action vector and to the
2229 * action trace list of all thread actions.
2231 * @param act is the ModelAction to add.
2233 void ModelExecution::add_action_to_lists(ModelAction *act)
2235 int tid = id_to_int(act->get_tid());
2236 ModelAction *uninit = NULL;
2238 action_list_t *list = get_safe_ptr_action(&obj_map, act->get_location());
2239 if (list->empty() && act->is_atomic_var()) {
2240 uninit = get_uninitialized_action(act);
2241 uninit_id = id_to_int(uninit->get_tid());
2242 list->push_front(uninit);
2244 list->push_back(act);
2246 action_trace.push_back(act);
2248 action_trace.push_front(uninit);
2250 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, act->get_location());
2251 if (tid >= (int)vec->size())
2252 vec->resize(priv->next_thread_id);
2253 (*vec)[tid].push_back(act);
2255 (*vec)[uninit_id].push_front(uninit);
2257 if ((int)thrd_last_action.size() <= tid)
2258 thrd_last_action.resize(get_num_threads());
2259 thrd_last_action[tid] = act;
2261 thrd_last_action[uninit_id] = uninit;
2263 if (act->is_fence() && act->is_release()) {
2264 if ((int)thrd_last_fence_release.size() <= tid)
2265 thrd_last_fence_release.resize(get_num_threads());
2266 thrd_last_fence_release[tid] = act;
2269 if (act->is_wait()) {
2270 void *mutex_loc = (void *) act->get_value();
2271 get_safe_ptr_action(&obj_map, mutex_loc)->push_back(act);
2273 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, mutex_loc);
2274 if (tid >= (int)vec->size())
2275 vec->resize(priv->next_thread_id);
2276 (*vec)[tid].push_back(act);
2281 * @brief Get the last action performed by a particular Thread
2282 * @param tid The thread ID of the Thread in question
2283 * @return The last action in the thread
2285 ModelAction * ModelExecution::get_last_action(thread_id_t tid) const
2287 int threadid = id_to_int(tid);
2288 if (threadid < (int)thrd_last_action.size())
2289 return thrd_last_action[id_to_int(tid)];
2295 * @brief Get the last fence release performed by a particular Thread
2296 * @param tid The thread ID of the Thread in question
2297 * @return The last fence release in the thread, if one exists; NULL otherwise
2299 ModelAction * ModelExecution::get_last_fence_release(thread_id_t tid) const
2301 int threadid = id_to_int(tid);
2302 if (threadid < (int)thrd_last_fence_release.size())
2303 return thrd_last_fence_release[id_to_int(tid)];
2309 * Gets the last memory_order_seq_cst write (in the total global sequence)
2310 * performed on a particular object (i.e., memory location), not including the
2312 * @param curr The current ModelAction; also denotes the object location to
2314 * @return The last seq_cst write
2316 ModelAction * ModelExecution::get_last_seq_cst_write(ModelAction *curr) const
2318 void *location = curr->get_location();
2319 action_list_t *list = obj_map.get(location);
2320 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2321 action_list_t::reverse_iterator rit;
2322 for (rit = list->rbegin(); (*rit) != curr; rit++)
2324 rit++; /* Skip past curr */
2325 for ( ; rit != list->rend(); rit++)
2326 if ((*rit)->is_write() && (*rit)->is_seqcst())
2332 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2333 * performed in a particular thread, prior to a particular fence.
2334 * @param tid The ID of the thread to check
2335 * @param before_fence The fence from which to begin the search; if NULL, then
2336 * search for the most recent fence in the thread.
2337 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2339 ModelAction * ModelExecution::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2341 /* All fences should have location FENCE_LOCATION */
2342 action_list_t *list = obj_map.get(FENCE_LOCATION);
2347 action_list_t::reverse_iterator rit = list->rbegin();
2350 for (; rit != list->rend(); rit++)
2351 if (*rit == before_fence)
2354 ASSERT(*rit == before_fence);
2358 for (; rit != list->rend(); rit++)
2359 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2365 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2366 * location). This function identifies the mutex according to the current
2367 * action, which is presumed to perform on the same mutex.
2368 * @param curr The current ModelAction; also denotes the object location to
2370 * @return The last unlock operation
2372 ModelAction * ModelExecution::get_last_unlock(ModelAction *curr) const
2374 void *location = curr->get_location();
2375 action_list_t *list = obj_map.get(location);
2376 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2377 action_list_t::reverse_iterator rit;
2378 for (rit = list->rbegin(); rit != list->rend(); rit++)
2379 if ((*rit)->is_unlock() || (*rit)->is_wait())
2384 ModelAction * ModelExecution::get_parent_action(thread_id_t tid) const
2386 ModelAction *parent = get_last_action(tid);
2388 parent = get_thread(tid)->get_creation();
2393 * Returns the clock vector for a given thread.
2394 * @param tid The thread whose clock vector we want
2395 * @return Desired clock vector
2397 ClockVector * ModelExecution::get_cv(thread_id_t tid) const
2399 return get_parent_action(tid)->get_cv();
2403 * @brief Find the promise (if any) to resolve for the current action and
2404 * remove it from the pending promise vector
2405 * @param curr The current ModelAction. Should be a write.
2406 * @return The Promise to resolve, if any; otherwise NULL
2408 Promise * ModelExecution::pop_promise_to_resolve(const ModelAction *curr)
2410 for (unsigned int i = 0; i < promises.size(); i++)
2411 if (curr->get_node()->get_promise(i)) {
2412 Promise *ret = promises[i];
2413 promises.erase(promises.begin() + i);
2420 * Resolve a Promise with a current write.
2421 * @param write The ModelAction that is fulfilling Promises
2422 * @param promise The Promise to resolve
2423 * @param work The work queue, for adding new fixup work
2424 * @return True if the Promise was successfully resolved; false otherwise
2426 bool ModelExecution::resolve_promise(ModelAction *write, Promise *promise,
2429 ModelVector<ModelAction *> actions_to_check;
2431 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2432 ModelAction *read = promise->get_reader(i);
2433 if (read_from(read, write)) {
2434 /* Propagate the changed clock vector */
2435 propagate_clockvector(read, work);
2437 actions_to_check.push_back(read);
2439 /* Make sure the promise's value matches the write's value */
2440 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2441 if (!mo_graph->resolvePromise(promise, write))
2442 priv->hard_failed_promise = true;
2445 * @todo It is possible to end up in an inconsistent state, where a
2446 * "resolved" promise may still be referenced if
2447 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2449 * Note that the inconsistency only matters when dumping mo_graph to
2455 //Check whether reading these writes has made threads unable to
2457 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2458 ModelAction *read = actions_to_check[i];
2459 mo_check_promises(read, true);
2466 * Compute the set of promises that could potentially be satisfied by this
2467 * action. Note that the set computation actually appears in the Node, not in
2469 * @param curr The ModelAction that may satisfy promises
2471 void ModelExecution::compute_promises(ModelAction *curr)
2473 for (unsigned int i = 0; i < promises.size(); i++) {
2474 Promise *promise = promises[i];
2475 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2478 bool satisfy = true;
2479 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2480 const ModelAction *act = promise->get_reader(j);
2481 if (act->happens_before(curr) ||
2482 act->could_synchronize_with(curr)) {
2488 curr->get_node()->set_promise(i);
2492 /** Checks promises in response to change in ClockVector Threads. */
2493 void ModelExecution::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2495 for (unsigned int i = 0; i < promises.size(); i++) {
2496 Promise *promise = promises[i];
2497 if (!promise->thread_is_available(tid))
2499 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2500 const ModelAction *act = promise->get_reader(j);
2501 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2502 merge_cv->synchronized_since(act)) {
2503 if (promise->eliminate_thread(tid)) {
2504 /* Promise has failed */
2505 priv->failed_promise = true;
2513 void ModelExecution::check_promises_thread_disabled()
2515 for (unsigned int i = 0; i < promises.size(); i++) {
2516 Promise *promise = promises[i];
2517 if (promise->has_failed()) {
2518 priv->failed_promise = true;
2525 * @brief Checks promises in response to addition to modification order for
2528 * We test whether threads are still available for satisfying promises after an
2529 * addition to our modification order constraints. Those that are unavailable
2530 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2531 * that promise has failed.
2533 * @param act The ModelAction which updated the modification order
2534 * @param is_read_check Should be true if act is a read and we must check for
2535 * updates to the store from which it read (there is a distinction here for
2536 * RMW's, which are both a load and a store)
2538 void ModelExecution::mo_check_promises(const ModelAction *act, bool is_read_check)
2540 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2542 for (unsigned int i = 0; i < promises.size(); i++) {
2543 Promise *promise = promises[i];
2545 // Is this promise on the same location?
2546 if (!promise->same_location(write))
2549 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2550 const ModelAction *pread = promise->get_reader(j);
2551 if (!pread->happens_before(act))
2553 if (mo_graph->checkPromise(write, promise)) {
2554 priv->hard_failed_promise = true;
2560 // Don't do any lookups twice for the same thread
2561 if (!promise->thread_is_available(act->get_tid()))
2564 if (mo_graph->checkReachable(promise, write)) {
2565 if (mo_graph->checkPromise(write, promise)) {
2566 priv->hard_failed_promise = true;
2574 * Compute the set of writes that may break the current pending release
2575 * sequence. This information is extracted from previou release sequence
2578 * @param curr The current ModelAction. Must be a release sequence fixup
2581 void ModelExecution::compute_relseq_breakwrites(ModelAction *curr)
2583 if (pending_rel_seqs.empty())
2586 struct release_seq *pending = pending_rel_seqs.back();
2587 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2588 const ModelAction *write = pending->writes[i];
2589 curr->get_node()->add_relseq_break(write);
2592 /* NULL means don't break the sequence; just synchronize */
2593 curr->get_node()->add_relseq_break(NULL);
2597 * Build up an initial set of all past writes that this 'read' action may read
2598 * from, as well as any previously-observed future values that must still be valid.
2600 * @param curr is the current ModelAction that we are exploring; it must be a
2603 void ModelExecution::build_may_read_from(ModelAction *curr)
2605 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
2607 ASSERT(curr->is_read());
2609 ModelAction *last_sc_write = NULL;
2611 if (curr->is_seqcst())
2612 last_sc_write = get_last_seq_cst_write(curr);
2614 /* Iterate over all threads */
2615 for (i = 0; i < thrd_lists->size(); i++) {
2616 /* Iterate over actions in thread, starting from most recent */
2617 action_list_t *list = &(*thrd_lists)[i];
2618 action_list_t::reverse_iterator rit;
2619 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2620 ModelAction *act = *rit;
2622 /* Only consider 'write' actions */
2623 if (!act->is_write() || act == curr)
2626 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2627 bool allow_read = true;
2629 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2631 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2635 /* Only add feasible reads */
2636 mo_graph->startChanges();
2637 r_modification_order(curr, act);
2638 if (!is_infeasible())
2639 curr->get_node()->add_read_from_past(act);
2640 mo_graph->rollbackChanges();
2643 /* Include at most one act per-thread that "happens before" curr */
2644 if (act->happens_before(curr))
2649 /* Inherit existing, promised future values */
2650 for (i = 0; i < promises.size(); i++) {
2651 const Promise *promise = promises[i];
2652 const ModelAction *promise_read = promise->get_reader(0);
2653 if (promise_read->same_var(curr)) {
2654 /* Only add feasible future-values */
2655 mo_graph->startChanges();
2656 r_modification_order(curr, promise);
2657 if (!is_infeasible())
2658 curr->get_node()->add_read_from_promise(promise_read);
2659 mo_graph->rollbackChanges();
2663 /* We may find no valid may-read-from only if the execution is doomed */
2664 if (!curr->get_node()->read_from_size()) {
2665 priv->no_valid_reads = true;
2669 if (DBG_ENABLED()) {
2670 model_print("Reached read action:\n");
2672 model_print("Printing read_from_past\n");
2673 curr->get_node()->print_read_from_past();
2674 model_print("End printing read_from_past\n");
2678 bool ModelExecution::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2680 for ( ; write != NULL; write = write->get_reads_from()) {
2681 /* UNINIT actions don't have a Node, and they never sleep */
2682 if (write->is_uninitialized())
2684 Node *prevnode = write->get_node()->get_parent();
2686 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2687 if (write->is_release() && thread_sleep)
2689 if (!write->is_rmw())
2696 * @brief Get an action representing an uninitialized atomic
2698 * This function may create a new one or try to retrieve one from the NodeStack
2700 * @param curr The current action, which prompts the creation of an UNINIT action
2701 * @return A pointer to the UNINIT ModelAction
2703 ModelAction * ModelExecution::get_uninitialized_action(const ModelAction *curr) const
2705 Node *node = curr->get_node();
2706 ModelAction *act = node->get_uninit_action();
2708 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), params->uninitvalue, model_thread);
2709 node->set_uninit_action(act);
2711 act->create_cv(NULL);
2715 static void print_list(const action_list_t *list)
2717 action_list_t::const_iterator it;
2719 model_print("------------------------------------------------------------------------------------\n");
2720 model_print("# t Action type MO Location Value Rf CV\n");
2721 model_print("------------------------------------------------------------------------------------\n");
2723 unsigned int hash = 0;
2725 for (it = list->begin(); it != list->end(); it++) {
2726 const ModelAction *act = *it;
2727 if (act->get_seq_number() > 0)
2729 hash = hash^(hash<<3)^((*it)->hash());
2731 model_print("HASH %u\n", hash);
2732 model_print("------------------------------------------------------------------------------------\n");
2735 #if SUPPORT_MOD_ORDER_DUMP
2736 void ModelExecution::dumpGraph(char *filename) const
2739 sprintf(buffer, "%s.dot", filename);
2740 FILE *file = fopen(buffer, "w");
2741 fprintf(file, "digraph %s {\n", filename);
2742 mo_graph->dumpNodes(file);
2743 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2745 for (action_list_t::const_iterator it = action_trace.begin(); it != action_trace.end(); it++) {
2746 ModelAction *act = *it;
2747 if (act->is_read()) {
2748 mo_graph->dot_print_node(file, act);
2749 if (act->get_reads_from())
2750 mo_graph->dot_print_edge(file,
2751 act->get_reads_from(),
2753 "label=\"rf\", color=red, weight=2");
2755 mo_graph->dot_print_edge(file,
2756 act->get_reads_from_promise(),
2758 "label=\"rf\", color=red");
2760 if (thread_array[act->get_tid()]) {
2761 mo_graph->dot_print_edge(file,
2762 thread_array[id_to_int(act->get_tid())],
2764 "label=\"sb\", color=blue, weight=400");
2767 thread_array[act->get_tid()] = act;
2769 fprintf(file, "}\n");
2770 model_free(thread_array);
2775 /** @brief Prints an execution trace summary. */
2776 void ModelExecution::print_summary() const
2778 #if SUPPORT_MOD_ORDER_DUMP
2779 char buffername[100];
2780 sprintf(buffername, "exec%04u", get_execution_number());
2781 mo_graph->dumpGraphToFile(buffername);
2782 sprintf(buffername, "graph%04u", get_execution_number());
2783 dumpGraph(buffername);
2786 model_print("Execution trace %d:", get_execution_number());
2787 if (isfeasibleprefix()) {
2788 if (is_yieldblocked())
2789 model_print(" YIELD BLOCKED");
2790 if (scheduler->all_threads_sleeping())
2791 model_print(" SLEEP-SET REDUNDANT");
2792 if (have_bug_reports())
2793 model_print(" DETECTED BUG(S)");
2795 print_infeasibility(" INFEASIBLE");
2798 print_list(&action_trace);
2801 if (!promises.empty()) {
2802 model_print("Pending promises:\n");
2803 for (unsigned int i = 0; i < promises.size(); i++) {
2804 model_print(" [P%u] ", i);
2805 promises[i]->print();
2812 * Add a Thread to the system for the first time. Should only be called once
2814 * @param t The Thread to add
2816 void ModelExecution::add_thread(Thread *t)
2818 unsigned int i = id_to_int(t->get_id());
2819 if (i >= thread_map.size())
2820 thread_map.resize(i + 1);
2822 if (!t->is_model_thread())
2823 scheduler->add_thread(t);
2827 * @brief Get a Thread reference by its ID
2828 * @param tid The Thread's ID
2829 * @return A Thread reference
2831 Thread * ModelExecution::get_thread(thread_id_t tid) const
2833 unsigned int i = id_to_int(tid);
2834 if (i < thread_map.size())
2835 return thread_map[i];
2840 * @brief Get a reference to the Thread in which a ModelAction was executed
2841 * @param act The ModelAction
2842 * @return A Thread reference
2844 Thread * ModelExecution::get_thread(const ModelAction *act) const
2846 return get_thread(act->get_tid());
2850 * @brief Get a Promise's "promise number"
2852 * A "promise number" is an index number that is unique to a promise, valid
2853 * only for a specific snapshot of an execution trace. Promises may come and go
2854 * as they are generated an resolved, so an index only retains meaning for the
2857 * @param promise The Promise to check
2858 * @return The promise index, if the promise still is valid; otherwise -1
2860 int ModelExecution::get_promise_number(const Promise *promise) const
2862 for (unsigned int i = 0; i < promises.size(); i++)
2863 if (promises[i] == promise)
2870 * @brief Check if a Thread is currently enabled
2871 * @param t The Thread to check
2872 * @return True if the Thread is currently enabled
2874 bool ModelExecution::is_enabled(Thread *t) const
2876 return scheduler->is_enabled(t);
2880 * @brief Check if a Thread is currently enabled
2881 * @param tid The ID of the Thread to check
2882 * @return True if the Thread is currently enabled
2884 bool ModelExecution::is_enabled(thread_id_t tid) const
2886 return scheduler->is_enabled(tid);
2890 * @brief Select the next thread to execute based on the curren action
2892 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
2893 * actions should be followed by the execution of their child thread. In either
2894 * case, the current action should determine the next thread schedule.
2896 * @param curr The current action
2897 * @return The next thread to run, if the current action will determine this
2898 * selection; otherwise NULL
2900 Thread * ModelExecution::action_select_next_thread(const ModelAction *curr) const
2902 /* Do not split atomic RMW */
2903 if (curr->is_rmwr())
2904 return get_thread(curr);
2905 if (curr->is_write()) {
2906 // std::memory_order order = curr->get_mo();
2908 // case std::memory_order_relaxed:
2909 // return get_thread(curr);
2910 // case std::memory_order_release:
2911 // return get_thread(curr);
2918 /* Follow CREATE with the created thread */
2919 /* which is not needed, because model.cc takes care of this */
2920 if (curr->get_type() == THREAD_CREATE)
2921 return curr->get_thread_operand();
2922 if (curr->get_type() == PTHREAD_CREATE) {
2923 return curr->get_thread_operand();
2928 /** @return True if the execution has taken too many steps */
2929 bool ModelExecution::too_many_steps() const
2931 return params->bound != 0 && priv->used_sequence_numbers > params->bound;
2935 * Takes the next step in the execution, if possible.
2936 * @param curr The current step to take
2937 * @return Returns the next Thread to run, if any; NULL if this execution
2940 Thread * ModelExecution::take_step(ModelAction *curr)
2942 Thread *curr_thrd = get_thread(curr);
2943 ASSERT(curr_thrd->get_state() == THREAD_READY);
2945 ASSERT(check_action_enabled(curr)); /* May have side effects? */
2946 curr = check_current_action(curr);
2949 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
2950 scheduler->remove_thread(curr_thrd);
2952 return action_select_next_thread(curr);
2956 * Launch end-of-execution release sequence fixups only when
2957 * the execution is otherwise feasible AND there are:
2959 * (1) pending release sequences
2960 * (2) pending assertions that could be invalidated by a change
2961 * in clock vectors (i.e., data races)
2962 * (3) no pending promises
2964 void ModelExecution::fixup_release_sequences()
2966 while (!pending_rel_seqs.empty() &&
2967 is_feasible_prefix_ignore_relseq() &&
2968 haveUnrealizedRaces()) {
2969 model_print("*** WARNING: release sequence fixup action "
2970 "(%zu pending release seuqence(s)) ***\n",
2971 pending_rel_seqs.size());
2972 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2973 std::memory_order_seq_cst, NULL, VALUE_NONE,